On 2014-01-31 12:32, Corrado Maurini wrote:
Le 31 janv. 2014 à 10:46, Anders Logg <[email protected]> a écrit :
I still don't see the point. How can you not accomplish that by
creating a subclass of NonlinearProblem directly, instead of via an
additional wrapper layer in the form of NonlinearDiscreteProblem?
Yes, I agree you can. But the codes (about 10 lines) would be always
the same in 90% of cases.
It sounds like what you are asking for is an interface in between
NonlinearProblem and NonlinearVariationalProblem, which takes care
of
assembly, but leaves nonlinear solve to the user. By the same
reasoning, we could add yet another interface which takes care of
nonlinear solve but leaves assembly to the user...
Yes this is exactly what of I was asking (and Patrick too, if I
correctly understood its message).
Let's say a default implementation of NonlinearProblem taking as input
something like NonlinearVariationalProblem.
But I understand that it is necessary and sane that you and Garth make
design choices to reduce the (too many) ways of formulating problems
and the interfaces to maintain.
After all it is really a minor point. It is reasonable that if someone
want to work at low level, he stays low-level both in assembling and
solving and takes care of copying and pasting its default
implementation of NonlinearProblem.
In any case the discussion was useful for me. That led me thinking
also that there may be some issues to care of about the way of
imposing non-homogenous boundary conditions when using
PETScSNESSolver. And the control of NonlinearProblem assembling could
be indeed useful in those cases.
Could someone register an Issue about the lack of symmetry that came up
earlier in the thread? Some recent changes in SystemAssembler now make
this an easy fix.
(Hopefully the last email in this this thread!)
Garth
Corrado
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
